On 2005-05-18, at 14:00:23 +0100, Dave Mitchell wrote:
> Based an a suggestion from Jarkko, I've tried building bleed with
>
> -Dregister=
>
> and timed 'make test' a few times.
> This is Linux FC3, GCC 3.4.3, Athlon 750Mhz, in single-user mode.
>
> There seems to be a slight advantage having the register keyword.
> Anyone want to test on other architectures/compilers?
$ uname -srmp
Linux 2.6.11.10 i686 Intel(R) Pentium(R) III Mobile CPU 1000MHz
$ gcc-4.1 --version
gcc-4.1 (GCC) 4.1.0 20050515 (experimental)
$ cat .patch
24501
threads
u=4.36 s=0.90 cu=383.57 cs=32.24 scripts=961 tests=107213
u=4.39 s=0.74 cu=389.46 cs=32.68 scripts=961 tests=107213
u=4.48 s=0.82 cu=392.08 cs=35.48 scripts=961 tests=107213
u=4.30 s=0.86 cu=386.84 cs=32.09 scripts=961 tests=107213
threads, -Dregister=
u=4.30 s=0.96 cu=386.94 cs=33.93 scripts=961 tests=107214
u=4.25 s=0.78 cu=393.28 cs=32.32 scripts=961 tests=107214
u=4.24 s=0.73 cu=392.16 cs=32.02 scripts=961 tests=107214
u=4.27 s=0.85 cu=401.33 cs=33.02 scripts=961 tests=107214
----
no threads
u=3.72 s=0.87 cu=313.80 cs=29.62 scripts=933 tests=106500
u=3.72 s=0.73 cu=319.16 cs=29.18 scripts=933 tests=106500
u=3.76 s=0.73 cu=319.63 cs=29.34 scripts=933 tests=106500
u=3.72 s=0.84 cu=319.21 cs=29.84 scripts=933 tests=106500
no threads, -Dregister=
u=3.77 s=0.83 cu=318.27 cs=29.86 scripts=933 tests=106501
u=3.62 s=0.67 cu=310.37 cs=29.42 scripts=933 tests=106501
u=3.68 s=0.74 cu=311.47 cs=29.23 scripts=933 tests=106501
u=3.94 s=0.93 cu=332.42 cs=31.98 scripts=933 tests=106501
--
The question of whether computers can think is just like the question of
whether submarines can swim.
-- Edsger W. Dijkstra
